llm evaluation

The shelf life of LLM findings

Authors

Owen Queen, James Zou

Venue

Zou Group, Stanford University

Published

2026-11-02

Owen Queen and James Zou revisit 50 published LLM findings across 20 models, using relative effect sizes and release-date trends to classify persistence. The interactive report presents methods and limitations and reports that 44% of tested claims expired and 34% weakened under its definitions.

Why I recommend it: The authors describe the work as preliminary; findings were not independently replicated for this listing.

zou-group.github.ioAdded Nov 9, 20260 opensCompare